Pollen season onset and allergen load tracking via vegetation phenology

Satellite vegetation indices from MODIS, Sentinel-2 and Landsat track canopy green-up timing as a proxy for pollen season onset. The signal is real but indirect: local calibration against pollen trap networks is required before regional allergen forecasts become actionable.

Sensors

MODIS MOD13 / MYD13 (Terra and Aqua): 250 m resolution NDVI and EVI composites at 16-day intervals; the foundational dataset for land surface phenology at continental scale. Terra and Aqua together reduce effective revisit to roughly 1–2 days for raw observations, though cloud contamination in spring limits usable composites. Archive runs from 2000, giving over two decades of seasonal baselines.
VIIRS VNP13 (Suomi NPP / NOAA-20): 500 m vegetation index product continuing the MODIS phenology record with improved radiometric calibration. 16-day composites; daily overpass frequency helps fill cloud gaps. Increasingly used to extend MODIS time series forward without discontinuity.
Sentinel-2 MSI (ESA): 10 m (visible and near-infrared) and 20 m (red-edge bands 5, 6, 7) resolution. The red-edge bands are particularly informative for chlorophyll content and early canopy flush, which MODIS cannot resolve. Revisit is 5 days at the equator with both satellites; cloud probability masks (Sen2Cor or s2cloudless) are essential in temperate spring conditions. Archive from 2015.
Landsat 8 / 9 OLI: 30 m multispectral bands including near-infrared for NDVI computation. 16-day revisit per satellite; combined 8-day revisit with both. Useful for constructing seasonal composites and detecting inter-annual shifts in green-up date at field scale. Archive back to 1972 (Landsat series) gives the longest phenological baseline of any operational system.

What the satellite actually measures, and what it does not

Every vegetation index time series records reflected solar radiation from a canopy. When chlorophyll concentrations rise in spring, red-band reflectance drops and near-infrared reflectance rises; NDVI and EVI climb accordingly. That climb is a reliable indicator of leaf-out timing. It is not a pollen measurement.

Pollen release is a physiological event that follows, or sometimes precedes, visible green-up depending on species. Wind-pollinated trees such as birch, alder and hazel shed pollen before or at bud-burst, meaning the satellite may detect canopy greenness only after the peak exposure event has already passed. Grasses and ragweed, by contrast, pollinate during or after canopy maturity. The satellite-to-pollen relationship is therefore species-specific, and applying a single phenological threshold across a mixed landscape will produce errors that only ground-truth data can quantify.

This is not a fatal limitation. It is a calibration problem. Published studies correlating MODIS green-up dates with Burkard pollen trap records across European and North American networks have shown statistically significant lead-lag relationships, typically within a window of 5 to 20 days depending on species and climate zone. That window is operationally useful for public health planning, provided the model is trained on local trap data and not borrowed from a different region.

Phenological metrics and how they are extracted

Land surface phenology (LSP) algorithms fit smooth curves to noisy vegetation index time series and extract transition dates: start of season (SOS), peak of season (POS), end of season (EOS) and the rate of green-up. TIMESAT, the MODIS Land Cover Dynamics product (MCD12Q2) and similar tools use double-logistic or Gaussian functions to do this. MCD12Q2 reports SOS and POS at 500 m resolution globally, with uncertainty estimates, and is freely available through NASA Earthdata.

Sentinel-2 adds spatial granularity that MODIS cannot provide. At 20 m in the red-edge bands, it is possible to distinguish phenological variation within a single woodland, separating early-flushing south-facing slopes from shaded north-facing stands. This matters for urban allergen mapping, where a city park and a street tree population may have meaningfully different green-up dates. The trade-off is that Sentinel-2's 5-day revisit, combined with spring cloud cover in temperate latitudes, frequently produces time series with gaps of 2 to 4 weeks. Gap-filling with MODIS or VIIRS data at coarser resolution is standard practice.

EVI is generally preferred over NDVI for dense canopies because it corrects for soil background and atmospheric effects, reducing saturation at high leaf area index values. For sparse or semi-arid vegetation relevant to weed pollen (ragweed in the Pannonian Basin, for example), NDVI performs adequately.

Building a regional pollen forecast from satellite phenology

The operational workflow has three stages. First, extract phenological metrics (SOS, POS, rate of green-up) for the species or land cover classes of interest across the study region. Second, correlate those metrics with historical pollen trap records to establish species-specific lag functions and concentration proxies. Third, issue seasonal outlooks when current-year green-up timing departs from the multi-year baseline.

A warm, early spring advances SOS. If the current-year SOS is 10 days earlier than the 20-year MODIS median, the birch pollen season is likely to start correspondingly early, and peak concentrations may be higher if the accelerated green-up is driven by sustained warmth rather than a brief warm spell followed by cold. That conditional logic requires meteorological co-variates; satellite phenology alone cannot distinguish the two.

Spatial resolution matters for health applications. A regional health authority needs district-level estimates, not continental averages. Sentinel-2 can provide 20 m green-up maps that, once calibrated, allow pollen load estimates at the scale of individual urban neighbourhoods. The honest caveat: calibration degrades rapidly with distance from the nearest pollen trap. A model trained on traps in one city should not be applied uncorrected to a city 300 km away with different species composition.

Honest limits: cloud, species ambiguity and the pollen trap gap

Cloud cover is the dominant operational constraint in temperate spring. Sentinel-2 may return no usable observations for 3 to 5 consecutive weeks over northern Europe in March and April, precisely the period when birch and alder phenology is most critical. MODIS compositing reduces but does not eliminate this problem. Radar satellites (Sentinel-1 SAR) can observe through cloud but do not provide vegetation index information directly; SAR-derived metrics such as backscatter change can supplement optical data for detecting canopy state transitions, though the relationship to pollen is even more indirect.

Species identification from space is imperfect. MODIS and even Sentinel-2 see a mixed canopy signal. Separating birch from oak in a temperate broadleaf forest requires either high-resolution hyperspectral data or a pre-existing species map derived from field survey or airborne data. Without species attribution, the phenological signal is an aggregate that may obscure the timing of the most allergenic taxon.

Ground-based pollen trap networks are sparse. In many countries, traps are concentrated in capital cities and university towns. Satellite phenology can extend spatial coverage, but the calibration is only as good as the nearest trap. For governments considering investment in allergen surveillance, the satellite layer is most valuable when it accompanies, not replaces, a minimum viable trap network.

From phenology map to public health output

The practical deliverables for a health ministry or national meteorological service are: a seasonal onset forecast issued 2 to 4 weeks before the expected pollen peak, based on current-year SOS anomaly relative to the multi-year baseline; a weekly or daily pollen load index map at district resolution during the active season; and an end-of-season summary comparing actual versus forecast timing and intensity for model refinement.

Satellize applies this workflow on open constellations, running phenological extraction on MODIS MOD13 and Sentinel-2 time series and correlating outputs with client-supplied or publicly available pollen trap records. The Tonga crop-estimation programme demonstrated the same core method: vegetation index time series calibrated against ground observations to produce actionable seasonal estimates. The pollen application differs in species specificity and the health-outcome framing, but the analytical architecture is the same.

For procurement teams: the satellite data are free (Copernicus, NASA Earthdata). The cost is in processing, calibration, validation and the operational infrastructure to issue timely forecasts. A pilot covering a single country's primary pollen season, with two or three dominant species, is a tractable starting point for assessing whether the satellite signal is strong enough in that climate and landscape to justify a full operational service.

Typical figures

Spatial resolution (phenology extraction)	250–500 m (MODIS/VIIRS); 20–30 m (Sentinel-2 red-edge, Landsat OLI)
Temporal resolution (composite period)	8–16 days for MODIS/VIIRS composites; 5-day revisit for Sentinel-2 (cloud-dependent)
Phenological metric latency	Near-real-time SOS detection possible within 1–2 weeks of green-up event; retrospective confirmation after EOS
Key spectral bands	Red (620–670 nm) and NIR (841–876 nm) for NDVI; red-edge (705–783 nm) on Sentinel-2 MSI for chlorophyll sensitivity
Archive depth	MODIS from 2000; Landsat series from 1972; Sentinel-2 from 2015; VIIRS VNP13 from 2012
Cloud limitation	Usable optical observations may be absent for 3–5 consecutive weeks in temperate spring; compositing and gap-filling required
Species discrimination	Land cover class level only without ancillary species maps; individual taxon separation requires hyperspectral or field-derived inputs
Pollen proxy accuracy	Species- and region-specific; published lag correlations typically r² 0.4–0.7 against trap records depending on taxon and climate zone
Coverage	Global (MODIS, VIIRS, Landsat); full Sentinel-2 coverage of Europe, much of Africa, Asia and the Americas
Delivery formats	GeoTIFF phenological metric layers; CSV seasonal onset tables; GIS-ready pollen index rasters; PDF seasonal outlook reports

Analytics Satellize can run

Green-up date anomaly map	Double-logistic curve fitting to MODIS MOD13 or VIIRS VNP13 NDVI/EVI time series; SOS extracted per pixel relative to 20-year baseline median	Annual GeoTIFF at 500 m showing days-earlier or days-later than baseline, by district; issued at season start
Sentinel-2 canopy flush progression layer	Red-edge chlorophyll index (CIre) time series at 20 m; cloud-masked with s2cloudless; smoothed with Savitzky-Golay filter to extract within-city green-up variability	Weekly GeoTIFF during active season; suitable for neighbourhood-level allergen risk communication
Species-class phenological calendar	Land cover stratification (e.g. ESA WorldCover or national forest inventory) applied to phenological metrics to extract SOS and POS per dominant species class	Tabular seasonal calendar by species class and administrative unit; updated annually
Pollen season onset forecast	Regression model linking current-year SOS anomaly and meteorological co-variates (accumulated growing degree days) to historical pollen trap peak dates; calibrated per taxon	Probabilistic onset forecast report issued 2–4 weeks before expected peak; includes confidence interval and comparison to prior 5-year range
In-season pollen load index	EVI rate-of-change during green-up correlated with concurrent trap-measured pollen concentration; index scaled to low/moderate/high/very high bands per published European pollen information standards	Daily or weekly district-level index map as GIS layer and PDF bulletin during active season
Inter-annual trend analysis	Mann-Kendall trend test on MODIS SOS time series 2000-present; identifies statistically significant advance or delay in green-up timing per land cover class	Decadal trend report with maps; supports long-term public health planning and climate adaptation evidence

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.